BRANCH 3
Movement
What connects chips that must share the work?
11 mapped nodes · binds in the optical path · the constraint is still moving upstream
Two facts from the output make this branch grow faster than the others. Models exceed one accelerator's memory, so a model that does not fit must be sharded and the shards exchange state for every token. And serving at scale means thousands of accelerators, where the fraction of time each spends waiting for the others is a major determinant of cluster efficiency.
Which produces the structural fact: interconnect scales faster than compute. Each generation raises the bandwidth every chip needs to reach every other chip, so optical demand grows faster than accelerator unit demand. Reasoning about optics from GPU shipments systematically undercounts—twice over, because lasers per module rise with lane count as well. The optical constraint has moved from module assembly to EML lasers and now toward InP substrates.
Optical transceiver modules
tight
Why it exists
The pluggable transceiver converts electrical signals into light and back. Volume is 800G today with 1.6T ramping and 3.2T next.
Chinese vendors hold seven of the top ten seats. Innolight and Eoptolink together supply roughly 60% of NVIDIA's 800G demand, Innolight alone above 50% of its 800G procurement.
The market is short: constrained by shortages of upstream EML laser chips and isolators, 1.6T shipments in 2026 are projected at about 15 million units against demand implying roughly 25 million — a gap of about 10 million units. LightCounting, April 2026.
Relevant businesses
- Innolight
- Eoptolink
- Coherent
- Lumentum
- Fabrinet
- AAOI
Worth understanding
Module assembly capacity is not what is short. The product is tight because laser chips and selected optical components cannot be bought in sufficient volume. This node names the finished transceiver, while the nodes feeding it identify where the scarcity actually lives.
↑ back to the map
Switch-ASIC wafer fabrication
available
Why it exists
A taped-out switch ASIC must be fabricated on a logic process suited to its die size, SerDes density, power and cost target before it can become switching silicon.
This is not the same node as accelerator wafer fabrication even when TSMC owns both factories. The products use separate masks, wafer starts, process choices and customer queues. A company spanning two verticals does not collapse their production paths into one node.
Relevant businesses
- TSMC
- Samsung Foundry
- Intel Foundry
Worth understanding
The shared dependency is one layer farther upstream: leading-edge logic fabs draw on some of the same EUV tool capacity. The dashed map edge shows that capability without pretending the accelerator and switch wafers are the same output.
↑ back to the map
Network switch ASICs
available
Why it exists
Something has to route the packets. Broadcom's Tomahawk and Jericho lead the merchant market; NVIDIA sells Spectrum-X and Quantum InfiniBand as part of an integrated system rather than as merchant silicon.
Not constrained. But this is where the architecture gets decided. Switch silicon dictates whether a link is copper or optical, at what rate, in what form factor. Small share of cost, enormous share of control.
Relevant businesses
- Broadcom
- NVIDIA
- Marvell
- Cisco
Worth understanding
NVIDIA's networking position is not really a networking business; it is a mechanism for selling the rack as a unit. When the compute vendor also supplies the fabric, every supplier below designs to a schedule set by someone else's product cycle.
↑ back to the map
Copper interconnects
available, physics-limited
Why it exists
There are two networks here with different physics. Scale-up connects accelerators inside a rack at the highest bandwidth over the shortest distance — NVIDIA's NVLink spine uses thousands of copper pairs. Scale-out connects racks across a building, which means optics.
Constrained by physics rather than capacity, and that boundary is the interesting part.
Relevant businesses
- Amphenol
- TE Connectivity
- Molex
- Luxshare
- Credo
Worth understanding
The copper/optics boundary is a moving line worth watching directly. Every generation raises the signalling rate, which shortens copper's usable reach. Simultaneously the scale-up domain grows larger, which lengthens the distances required. Those two trends point in opposite directions, and the point where copper stops working is where optics starts getting paid.
Copper as a material is also shared with branch 4, where it goes into transformer windings and busway. Two branches, one metal.
↑ back to the map
Low-loss laminates
tight
Why it exists
At very high signalling rates, ordinary board material loses too much energy and degrades the signal. Low-loss copper-clad laminate is the substrate from which the network PCB is fabricated.
The specialty material is tighter than generic board capacity. Separating it from the PCB identifies the upstream line whose formulation, qualification and expansion clock actually limits the highest-performance builds.
Relevant businesses
- Elite Material
- Rogers
- Doosan
Worth understanding
This is a quiet second-order constraint that appears in material lead times before it appears in system shipments, making it a useful early-warning instrument for the branch.
↑ back to the map
High-speed network PCBs
tight
Why it exists
Switch ASICs, power delivery, connectors and optics must be assembled on very high-layer-count boards that preserve signal integrity at network speed.
Selectively tight at the top end. The board shop must control registration, vias, impedance and yield across layer counts far beyond ordinary electronics; it is a fabrication capability distinct from the laminate it consumes.
Relevant businesses
Worth understanding
Keeping board fabrication separate from laminate supply makes the release valves legible: more material does not create more qualified board capacity, and more board lines do not help if the qualified material is short.
↑ back to the map
NICs, DPUs and network adapters
available
Why it exists
A switch and a cable do not connect a server by themselves. Each compute endpoint needs an adapter that terminates the link, moves data to and from the accelerator or host, and increasingly offloads transport, security and storage work.
This node joins at rack integration rather than inside the switch system. It belongs to the server endpoint, while switch ASICs, boards and link hardware form the other side of the network.
Relevant businesses
- NVIDIA
- Broadcom
- Marvell
- Intel
- AMD/Pensando
Worth understanding
NIC and DPU content rises with cluster scale and architecture complexity even when accelerator count is unchanged. Omitting the endpoint makes the map look as though a network fabric can attach itself directly to a GPU package.
↑ back to the map
Optical-module electronics
tight
Why it exists
The DSP, laser driver and receiver electronics condition, encode and recover the electrical signal on either side of the optical path.
Selected devices are tight, but they are not the same production problem as laser chips or module assembly. Optical passives and isolators remain inside the module detail at this map resolution rather than being promoted into ceremonial nodes without independent economics.
Relevant businesses
- Marvell
- Broadcom
- MACOM
- Semtech
↑ back to the map
EML laser chips
binding
Why it exists
Something has to make the light. For 800G and 1.6T datacentre links that means electro-absorption modulated lasers, one per lane. A 1.6T module at 200G per lane needs eight of them.
This is where the optical bottleneck actually lives, and it is tightening rather than easing. Lumentum is currently the only supplier shipping 200G-per-lane EMLs at volume — the exact part 1.6T requires. EE Times Asia. Lead times extend beyond 2027, and NVIDIA committed roughly $4bn to Lumentum and Coherent, locking supply and pushing rivals past 2027. reported May 2026. Prepayments from Innolight, Eoptolink and TFC rose from ¥172m to ¥2.27bn in Q1 2026, a thirteen-fold increase.
Relevant businesses
- Lumentum
- Coherent
- Mitsubishi
- Sumitomo
- AAOI (captive)
Worth understanding
Two signals, from opposite sides of the market, saying the same thing. A thirteen-fold rise in prepayments is a price signal from people with better information than the market — that is what conviction about scarcity looks like on a balance sheet. And a customer does not invest four billion dollars in two component suppliers to secure a component it expects to be plentiful.
Then the mechanism that matters most. Vertical integration inverts in value when the input is what is scarce. Owning laser fabs is a capital burden and a margin drag when lasers are abundant — the ordinary state, and the one most models are calibrated on. When lasers are the constraint, in-house supply is not a cost centre; it is an allocation that competitors are queuing and prepaying years early to obtain.
The inversion is not permanent, and its temporariness is the risk. The question is duration, and duration is set by epitaxy and substrate lead times — not by anything the module vendors control.
↑ back to the map
Indium-phosphide substrates
bedrock
Why it exists
EML lasers are grown on indium phosphide. Supply is concentrated in two primary names, AXT and Sumitomo Electric, with JX as an adjacent third supplier.
Indium phosphide shortage has emerged as the AI optical interconnect bottleneck TrendForce, 2026-08-06. That is the constraint arriving at the floor of this branch.
Relevant businesses
Worth understanding
Material constraints behave differently from capacity constraints. A fab takes two to three years. An epitaxy line takes longer. A new compound-substrate supplier needs a decade of process learning and customer qualification. When a material binds, it binds for longer than anything downstream of it — and the present map does not carry an investable layer farther upstream.
Note what the sequence says. The constraint on this branch has climbed twice in about two years: module assembly, then laser chip, now substrate. Each move went closer to the raw material, into a narrower supplier set, with a longer expansion lead time. That is the opposite direction from branch 1, and it relocates the value: when a constraint moves downstream, value goes to whoever integrates; when it climbs, value goes to whoever owns the input.
↑ back to the map
Network switch and interconnect system
tight
Why it exists
Switch ASICs, qualified boards, optical or copper links and control software must become working switch and interconnect hardware before racks can be connected into a cluster.
This is deliberately not called “network fabric.” A fabric can mean the already-deployed topology. The node here is the tangible switch-and-link system that arrives at rack integration; cluster configuration and commissioning happen after those physical products meet the servers.
Relevant businesses
- NVIDIA
- Broadcom ecosystem
- Arista
- Cisco and system partners
Worth understanding
Its state inherits the scarcest component path. Available switch silicon cannot create a complete system when transceiver supply is short, which is why the merge is economically informative rather than a decorative box.
↑ back to the map
The threat that does not come from below: co-packaged optics
Not a node on this branch — a demand-side release valve on it. Instead of a pluggable module at the switch faceplate, the optical engine is integrated into the switch package beside the ASIC. NVIDIA ships Quantum-X InfiniBand in early 2026 and Spectrum-X Photonics Ethernet in the second half NVIDIA, claiming 4x fewer lasers, 3.5x better power efficiency, and roughly 40MW saved in a large deployment.
“4x fewer lasers” is the headline that matters to anyone holding optical exposure, and it needs care in three directions. It applies to the switch side of the link, not the whole link — the accelerator end and the enormous installed base of pluggable ports do not disappear. NVIDIA's own announcement names pluggable vendors as continuing partners: Coherent, Eoptolink, Fabrinet and Innolight appear in the same release. And fewer lasers per port is not fewer lasers — demand scales as ports × lasers-per-port, so if ports grow faster than 4x over the transition, aggregate laser demand still rises. Whether CPO reduces total laser demand is an arithmetic question about two growth rates, not a qualitative one, and most commentary treats it as qualitative.
Silicon photonics is the more consequential version of the same question, because it replaces per-lane EMLs with a smaller number of external light sources. Both are real, both move slower than their announcements imply, and neither is settled inside 2027.
Reading this branch
The constraint has moved upstream twice in about two years: finished transceiver availability → EML laser chip → InP substrate. Each move went closer to the material, into a narrower supplier set, with a longer expansion lead time. A reader who learned the shape in 2024 is watching the wrong node.
Upward migration relocates the value. When a constraint moves downstream, value goes to whoever integrates the scarce combination. When it climbs, value goes to whoever owns the input. That is why the direction of travel matters and not merely the fact of movement.
The genuine risk is not competition — it is architecture. Nothing here is threatened by another module vendor. It is threatened by CPO and, more seriously, by silicon photonics removing the need for per-lane EMLs. That risk resolves on an engineering schedule rather than a capacity one, which makes it harder to forecast and easier to be blindsided by.